* doc/inference api (#11332)

* inference API init cn (#11731)

* hack deadline to fix timeout

* add fixme

Fix dist train broadcasting bug

fix pserver sub-blocks

fix sparse paraexe dist train

Fix async update failed

* 1. Remove PYTHON_FLAGS from paddle_build.sh in paddlepaddle/paddle:latest-dev

* 1. Add PYTHON_FLAGS back
2. Change SIGKILL to SIGINT and SIGTERM

* 1. Add setup.py.in back

* 1. add pip install open-cv in Dockerfile to avoid libusb_exit hanging up which is caused by the opencv-python package missing

* 1. Add the && \ to line above

* 1. Remove the notice comment

* Fix Parallel Exe(VarHandel's version)

* Fix broadcast

* enhance ParallelExecutor stable

disable the LODTensor warning for now

small thread-safety fix and doc improvements.

* Make the CUDA kernel of concat correct and fix unit tests.

* Add sub_blocks of lr_decay_op to pserver_prog after distribute_transpiler

* Remove unused logs and logics

* 1. Add ops to new block (considering the nested block condition)
2. Follow the original hierarchy of blocks
3. Change the function's name and remove debug lines

* Remove tape

* remove tape in cmake

* fix CI